ofiere-openclaw-plugin 4.56.8 → 4.56.10

package/README.md

@@ -1,104 +1,104 @@
-# Ofiere PM Plugin for OpenClaw
-
-Manage your Ofiere PM dashboard directly from OpenClaw agents. Create tasks, manage projects, build workflows, store knowledge — all synced to the dashboard in real time.
-
-## Quick Install (One-Click)
-
-```bash
-curl -sSL https://ofiere.com/scripts/install.sh | bash -s -- \
-  --supabase-url "https://xxx.supabase.co" \
-  --service-key "eyJ..." \
-  --user-id "your-uuid"
-```
-
-Only 3 parameters needed. All agents get the plugin automatically.
-
-## Uninstall
-
-```bash
-curl -sSL https://ofiere.com/scripts/uninstall.sh | bash
-```
-
-## How It Works
-
-Once configured, the plugin connects to your Supabase database at gateway startup and registers 9 PM meta-tools directly into the agent. There's no separate MCP server process — it runs inside the OpenClaw gateway for maximum reliability.
-
-Changes made by the agent are immediately visible on the Ofiere dashboard (Vercel) via Supabase real-time subscriptions.
-
-## AI Meta-Tools
-
-The plugin uses a scalable meta-tool architecture. Each tool handles one domain with an `action` parameter to select the operation.
-
-| Tool | Actions | Description |
-|---|---|---|
-| `OFIERE_TASK_OPS` | `list`, `create`, `update`, `delete` | Rich task management with execution plans, goals, constraints |
-| `OFIERE_AGENT_OPS` | `list` | Query available agents for task assignment |
-| `OFIERE_PROJECT_OPS` | `list_spaces`, `create_space`, `create_folder`, `bulk_create_tasks`, etc. | Full project hierarchy: spaces → folders → tasks |
-| `OFIERE_SCHEDULE_OPS` | `list`, `create`, `update`, `delete` | Calendar events with recurrence |
-| `OFIERE_KNOWLEDGE_OPS` | `search`, `list`, `create`, `update`, `delete` | Knowledge library with full content retrieval |
-| `OFIERE_WORKFLOW_OPS` | `list`, `get`, `create`, `update`, `delete`, `list_runs`, `trigger` | Visual workflow builder with 16 node types |
-| `OFIERE_NOTIFY_OPS` | `list`, `mark_read`, `create` | In-app notifications |
-| `OFIERE_MEMORY_OPS` | `list_conversations`, `search_knowledge` | Conversation history & agent memory |
-| `OFIERE_PROMPT_OPS` | `list`, `get`, `create`, `update`, `delete` | System prompt chunk management |
-
-### Example
-
-```
-// Create a task with execution plan
-OFIERE_TASK_OPS({ action: "create", title: "Deploy v2", agent_id: "ivy",
-  execution_plan: [{ step: 1, action: "Build", detail: "Run production build" }] })
-
-// Create a workflow with nodes
-OFIERE_WORKFLOW_OPS({ action: "create", name: "Deploy Pipeline",
-  nodes: [
-    { type: "agent_step", data: { label: "Build", task: "Run npm build" } },
-    { type: "human_approval", data: { label: "Review", instructions: "Check build output" } },
-    { type: "output", data: { label: "Done" } }
-  ],
-  edges: [
-    { source: "agent_step-...", target: "human_approval-..." },
-    { source: "human_approval-...", target: "output-..." }
-  ]
-})
-
-// Search knowledge library
-OFIERE_KNOWLEDGE_OPS({ action: "search", query: "API rate limits" })
-```
-
-## CLI Commands
-
-```bash
-openclaw ofiere setup     # Configure Supabase connection and agent identity
-openclaw ofiere status    # View current configuration
-openclaw ofiere doctor    # Test connection and list agents
-```
-
-## Configuration
-
-Set via `openclaw ofiere setup` or environment variables:
-
-| Option | Env Var | Description |
-|---|---|---|
-| `supabaseUrl` | `OFIERE_SUPABASE_URL` | Supabase project URL |
-| `serviceRoleKey` | `OFIERE_SERVICE_ROLE_KEY` | Supabase service role key |
-| `userId` | `OFIERE_USER_ID` | Your user UUID |
-| `agentId` | `OFIERE_AGENT_ID` | This agent's ID (optional — auto-detected) |
-| `enabled` | — | Enable/disable the plugin (default: `true`) |
-
-## Architecture
-
-```
-OpenClaw Agent (VPS)
-     │ plugin runs IN-PROCESS
-Ofiere Plugin ──► Supabase (shared database)
-                      ▲
-Ofiere Dashboard ─────┘  (Vercel, real-time)
-```
-
-Both the agent plugin and the Vercel dashboard talk to the same Supabase instance. When the agent creates/updates a task, the dashboard sees it instantly through Supabase real-time subscriptions.
-
-## Links
-
-- [Ofiere Dashboard](https://github.com/gilanggemar/Ofiere)
-- [OpenClaw](https://openclaw.ai)
-- [Supabase](https://supabase.com)
+# Ofiere PM Plugin for OpenClaw
+
+Manage your Ofiere PM dashboard directly from OpenClaw agents. Create tasks, manage projects, build workflows, store knowledge — all synced to the dashboard in real time.
+
+## Quick Install (One-Click)
+
+```bash
+curl -sSL https://ofiere.com/scripts/install.sh | bash -s -- \
+  --supabase-url "https://xxx.supabase.co" \
+  --service-key "eyJ..." \
+  --user-id "your-uuid"
+```
+
+Only 3 parameters needed. All agents get the plugin automatically.
+
+## Uninstall
+
+```bash
+curl -sSL https://ofiere.com/scripts/uninstall.sh | bash
+```
+
+## How It Works
+
+Once configured, the plugin connects to your Supabase database at gateway startup and registers 9 PM meta-tools directly into the agent. There's no separate MCP server process — it runs inside the OpenClaw gateway for maximum reliability.
+
+Changes made by the agent are immediately visible on the Ofiere dashboard (Vercel) via Supabase real-time subscriptions.
+
+## AI Meta-Tools
+
+The plugin uses a scalable meta-tool architecture. Each tool handles one domain with an `action` parameter to select the operation.
+
+| Tool | Actions | Description |
+|---|---|---|
+| `OFIERE_TASK_OPS` | `list`, `create`, `update`, `delete` | Rich task management with execution plans, goals, constraints |
+| `OFIERE_AGENT_OPS` | `list` | Query available agents for task assignment |
+| `OFIERE_PROJECT_OPS` | `list_spaces`, `create_space`, `create_folder`, `bulk_create_tasks`, etc. | Full project hierarchy: spaces → folders → tasks |
+| `OFIERE_SCHEDULE_OPS` | `list`, `create`, `update`, `delete` | Calendar events with recurrence |
+| `OFIERE_KNOWLEDGE_OPS` | `search`, `list`, `create`, `update`, `delete` | Knowledge library with full content retrieval |
+| `OFIERE_WORKFLOW_OPS` | `list`, `get`, `create`, `update`, `delete`, `list_runs`, `trigger` | Visual workflow builder with 16 node types |
+| `OFIERE_NOTIFY_OPS` | `list`, `mark_read`, `create` | In-app notifications |
+| `OFIERE_MEMORY_OPS` | `list_conversations`, `search_knowledge` | Conversation history & agent memory |
+| `OFIERE_PROMPT_OPS` | `list`, `get`, `create`, `update`, `delete` | System prompt chunk management |
+
+### Example
+
+```
+// Create a task with execution plan
+OFIERE_TASK_OPS({ action: "create", title: "Deploy v2", agent_id: "ivy",
+  execution_plan: [{ step: 1, action: "Build", detail: "Run production build" }] })
+
+// Create a workflow with nodes
+OFIERE_WORKFLOW_OPS({ action: "create", name: "Deploy Pipeline",
+  nodes: [
+    { type: "agent_step", data: { label: "Build", task: "Run npm build" } },
+    { type: "human_approval", data: { label: "Review", instructions: "Check build output" } },
+    { type: "output", data: { label: "Done" } }
+  ],
+  edges: [
+    { source: "agent_step-...", target: "human_approval-..." },
+    { source: "human_approval-...", target: "output-..." }
+  ]
+})
+
+// Search knowledge library
+OFIERE_KNOWLEDGE_OPS({ action: "search", query: "API rate limits" })
+```
+
+## CLI Commands
+
+```bash
+openclaw ofiere setup     # Configure Supabase connection and agent identity
+openclaw ofiere status    # View current configuration
+openclaw ofiere doctor    # Test connection and list agents
+```
+
+## Configuration
+
+Set via `openclaw ofiere setup` or environment variables:
+
+| Option | Env Var | Description |
+|---|---|---|
+| `supabaseUrl` | `OFIERE_SUPABASE_URL` | Supabase project URL |
+| `serviceRoleKey` | `OFIERE_SERVICE_ROLE_KEY` | Supabase service role key |
+| `userId` | `OFIERE_USER_ID` | Your user UUID |
+| `agentId` | `OFIERE_AGENT_ID` | This agent's ID (optional — auto-detected) |
+| `enabled` | — | Enable/disable the plugin (default: `true`) |
+
+## Architecture
+
+```
+OpenClaw Agent (VPS)
+     │ plugin runs IN-PROCESS
+Ofiere Plugin ──► Supabase (shared database)
+                      ▲
+Ofiere Dashboard ─────┘  (Vercel, real-time)
+```
+
+Both the agent plugin and the Vercel dashboard talk to the same Supabase instance. When the agent creates/updates a task, the dashboard sees it instantly through Supabase real-time subscriptions.
+
+## Links
+
+- [Ofiere Dashboard](https://github.com/gilanggemar/Ofiere)
+- [OpenClaw](https://openclaw.ai)
+- [Supabase](https://supabase.com)

package/dist/src/prompt.js

@@ -31,92 +31,92 @@ const TOOL_SUMMARIES = {
 // These are used as tool descriptions in registerTools(), NOT in the system prompt.
 // The LLM sees them only when exploring available tools or preparing a tool call.
 export const TOOL_DOCS = {
-    OFIERE_TASK_OPS: `Manage tasks and approvals.
-Actions: "list", "create", "update", "delete", "add_approval", "list_approvals", "resolve_approval"
-- list: Filter by status, agent_id, space_id, folder_id, limit. Returns execution_plan, goals, constraints if present
-- create: Requires title. IMPORTANT: Always pass agent_id with your own name to self-assign (e.g. agent_id: "celia")
-  - Optional: description, instructions, execution_plan, goals, constraints, system_prompt, priority, tags, dates
-  - For COMPLEX tasks: include execution_plan (step-by-step), goals, constraints, and system_prompt
-  - For SIMPLE tasks: just title and optionally description
-- update: Requires task_id. All create fields + progress
-- delete: Requires task_id. Removes task and subtasks
-- add_approval: Request sign-off. Requires: task_id, approver_name. Optional: approver_type (human|agent), due_date, comment
-- list_approvals: List approvals. Optional: task_id, approval_status (pending|approved|rejected)
+    OFIERE_TASK_OPS: `Manage tasks and approvals.
+Actions: "list", "create", "update", "delete", "add_approval", "list_approvals", "resolve_approval"
+- list: Filter by status, agent_id, space_id, folder_id, limit. Returns execution_plan, goals, constraints if present
+- create: Requires title. IMPORTANT: Always pass agent_id with your own name to self-assign (e.g. agent_id: "celia")
+  - Optional: description, instructions, execution_plan, goals, constraints, system_prompt, priority, tags, dates
+  - For COMPLEX tasks: include execution_plan (step-by-step), goals, constraints, and system_prompt
+  - For SIMPLE tasks: just title and optionally description
+- update: Requires task_id. All create fields + progress
+- delete: Requires task_id. Removes task and subtasks
+- add_approval: Request sign-off. Requires: task_id, approver_name. Optional: approver_type (human|agent), due_date, comment
+- list_approvals: List approvals. Optional: task_id, approval_status (pending|approved|rejected)
 - resolve_approval: Approve/reject. Requires: approval_id, approval_status. Optional: comment`,
-    OFIERE_AGENT_OPS: `Query + manage agents and their staff subagents.
-Actions: "list", "list_subagents", "create_subagent", "delete_subagent", "invalidate_tier_cache"
-- list: All top-level agents (chiefs / native + OpenClaw) with IDs, names, roles. Use for task assignment lookup.
-- list_subagents: Staff under a chief. Required: chief_agent_id.
-- create_subagent: Add a staff subagent under a chief (max 5 per chief). Required: chief_agent_id, name. Optional: role (default "Staff"), codename, color_hex.
-- delete_subagent: Remove a staff subagent. Required: subagent_id.
+    OFIERE_AGENT_OPS: `Query + manage agents and their staff subagents.
+Actions: "list", "list_subagents", "create_subagent", "delete_subagent", "invalidate_tier_cache"
+- list: All top-level agents (chiefs / native + OpenClaw) with IDs, names, roles. Use for task assignment lookup.
+- list_subagents: Staff under a chief. Required: chief_agent_id.
+- create_subagent: Add a staff subagent under a chief (max 5 per chief). Required: chief_agent_id, name. Optional: role (default "Staff"), codename, color_hex.
+- delete_subagent: Remove a staff subagent. Required: subagent_id.
 - invalidate_tier_cache: Flush plugin's in-process tier cache (5 min TTL). Optional: agent_id. Use after direct-DB mutation of agent_tier_overrides.`,
-    OFIERE_PROJECT_OPS: `Manage PM hierarchy.
-Actions: "list_spaces", "create_space", "update_space", "delete_space", "list_folders", "create_folder", "update_folder", "delete_folder", "list_dependencies", "add_dependency", "remove_dependency"
-- Spaces: Top-level containers. CRUD with name, icon, icon_color
-- Folders: Inside spaces. Nest via parent_folder_id. Types: folder, project
+    OFIERE_PROJECT_OPS: `Manage PM hierarchy.
+Actions: "list_spaces", "create_space", "update_space", "delete_space", "list_folders", "create_folder", "update_folder", "delete_folder", "list_dependencies", "add_dependency", "remove_dependency"
+- Spaces: Top-level containers. CRUD with name, icon, icon_color
+- Folders: Inside spaces. Nest via parent_folder_id. Types: folder, project
 - Dependencies: Link tasks. Types: finish_to_start (default), start_to_start, finish_to_finish, start_to_finish. lag_days optional`,
-    OFIERE_SCHEDULE_OPS: `Calendar events.
-Actions: "list", "create", "update", "delete"
-- list: Filter by start_date, end_date, agent_id
-- create: Requires title + scheduled_date (YYYY-MM-DD). Optional: scheduled_time, duration_minutes, task_id, agent_id, recurrence_type, color
+    OFIERE_SCHEDULE_OPS: `Calendar events.
+Actions: "list", "create", "update", "delete"
+- list: Filter by start_date, end_date, agent_id
+- create: Requires title + scheduled_date (YYYY-MM-DD). Optional: scheduled_time, duration_minutes, task_id, agent_id, recurrence_type, color
 - Recurrence: none, hourly, daily, weekly, monthly with interval`,
-    OFIERE_KNOWLEDGE_OPS: `Ofiere Knowledge Library.
-Actions: "search", "list", "create", "update", "delete"
-- ALWAYS use when user mentions "knowledge base", "knowledge library", or asks to recall stored knowledge
-- search: Keyword search. Requires: query
-- list: Paginated listing with full content. Optional: search filter
-- create: Requires: file_name. Optional: content, source, author, credibility_tier
+    OFIERE_KNOWLEDGE_OPS: `Ofiere Knowledge Library.
+Actions: "search", "list", "create", "update", "delete"
+- ALWAYS use when user mentions "knowledge base", "knowledge library", or asks to recall stored knowledge
+- search: Keyword search. Requires: query
+- list: Paginated listing with full content. Optional: search filter
+- create: Requires: file_name. Optional: content, source, author, credibility_tier
 - update/delete: By document ID`,
-    OFIERE_WORKFLOW_OPS: `Full workflow automation (13 actions).
-Actions: "list", "get", "create", "update", "delete", "list_runs", "trigger", "add_nodes", "update_node", "delete_nodes", "add_edges", "delete_edges", "insert_node_between"
-- ALWAYS call "get" before modifying to see node IDs and graph structure
-- Node types: manual_trigger, webhook_trigger, agent_step, formatter_step, http_request, task_call, variable_set, condition, human_approval, delay, loop, convergence, output, checkpoint, note
-- Edge handles: condition edges use "condition-true"/"condition-false". Loop edges use "loop_body"/"done"
+    OFIERE_WORKFLOW_OPS: `Full workflow automation (13 actions).
+Actions: "list", "get", "create", "update", "delete", "list_runs", "trigger", "add_nodes", "update_node", "delete_nodes", "add_edges", "delete_edges", "insert_node_between"
+- ALWAYS call "get" before modifying to see node IDs and graph structure
+- Node types: manual_trigger, webhook_trigger, agent_step, formatter_step, http_request, task_call, variable_set, condition, human_approval, delay, loop, convergence, output, checkpoint, note
+- Edge handles: condition edges use "condition-true"/"condition-false". Loop edges use "loop_body"/"done"
 - Variables: {{prev.nodeId.outputText}}, {{variables.key}}`,
-    OFIERE_NOTIFY_OPS: `Notifications & Channel Reports.
-Actions: "list", "mark_read", "mark_all_read", "delete", "send_report", "get_task_detail", "schedule_report", "list_schedules", "delete_schedule"
-- send_report: Send PM progress report to YOUR channels. Required: scope_type, agent_id (YOUR name)
-- get_task_detail: Get full task result by report number. Required: task_number
+    OFIERE_NOTIFY_OPS: `Notifications & Channel Reports.
+Actions: "list", "mark_read", "mark_all_read", "delete", "send_report", "get_task_detail", "schedule_report", "list_schedules", "delete_schedule"
+- send_report: Send PM progress report to YOUR channels. Required: scope_type, agent_id (YOUR name)
+- get_task_detail: Get full task result by report number. Required: task_number
 - schedule_report: Recurring reports. Required: scope_type, recurrence_type, agent_id`,
-    OFIERE_MEMORY_OPS: `Conversation history & knowledge memory.
+    OFIERE_MEMORY_OPS: `Conversation history & knowledge memory.
 Actions: "list_conversations", "get_messages", "search_messages", "add_knowledge", "search_knowledge"`,
-    OFIERE_PROMPT_OPS: `Manage prompt instruction chunks.
-Actions: "list", "get", "create", "update", "delete"
+    OFIERE_PROMPT_OPS: `Manage prompt instruction chunks.
+Actions: "list", "get", "create", "update", "delete"
 - create: New chunk with name (max 30 chars) + content. Optional: color (hex), category`,
-    OFIERE_CONSTELLATION_OPS: `Create/edit/delete OpenClaw agents from chat.
-Actions: "list_agents", "get_agent", "read_file", "write_file", "create_agent", "delete_agent", "delete_file", "read_blueprint", "list_agent_mesh"
-- CRITICAL: ALWAYS call read_blueprint before creating/editing agents
+    OFIERE_CONSTELLATION_OPS: `Create/edit/delete OpenClaw agents from chat.
+Actions: "list_agents", "get_agent", "read_file", "write_file", "create_agent", "delete_agent", "delete_file", "read_blueprint", "list_agent_mesh"
+- CRITICAL: ALWAYS call read_blueprint before creating/editing agents
 - delete_agent: IRREVERSIBLE. Always ask user for explicit confirmation first`,
-    OFIERE_FILE_OPS: `Manage Space Files.
-Actions: "list_files", "list_folders", "create_folder", "create_text_file", "upload_file", "read_text_file", "rename_file", "rename_folder", "move_file", "move_folder", "delete_file", "delete_folder", "share_file", "unshare_file"
-- read_text_file: Use when task references @[name](file:ID)
+    OFIERE_FILE_OPS: `Manage Space Files.
+Actions: "list_files", "list_folders", "create_folder", "create_text_file", "upload_file", "read_text_file", "rename_file", "rename_folder", "move_file", "move_folder", "delete_file", "delete_folder", "share_file", "unshare_file"
+- read_text_file: Use when task references @[name](file:ID)
 - Files appear in PM Space Files tab immediately`,
-    OFIERE_PLAN_OPS: `Visual execution plan builder.
-Actions: "list", "get", "create", "update", "delete", "add_nodes", "execute"
-- Plans are visual DAG drafts — not real tasks until "execute"
-- Node types: task, gate, milestone
+    OFIERE_PLAN_OPS: `Visual execution plan builder.
+Actions: "list", "get", "create", "update", "delete", "add_nodes", "execute"
+- Plans are visual DAG drafts — not real tasks until "execute"
+- Node types: task, gate, milestone
 - Execution maps ALL fields: execution_steps, goals, constraints, system_prompt`,
-    OFIERE_SOP_OPS: `Standard Operating Procedures.
-Actions: "list_templates", "create", "list", "get", "update", "delete", "list_subagents", "apply_template", "propose_attach", "commit_attach"
-- sop_data: { title, purpose, scope, applicability, prerequisites:{ conditions, required_tools, required_permissions, safety_warnings }, procedure_steps[], expected_outputs[], escalation_rules[], acceptance_criteria[], rollback_procedure, notes }
-- Legacy field names still accepted (objective→purpose, steps→procedure_steps, deliverables→expected_outputs, escalationRules→escalation_rules, successCriteria→acceptance_criteria) — prefer new names.
-- propose_attach / commit_attach: attach SOPs to a run (target_kind: conversation|task|scheduler_event). Tier rule: SOPs only attach to STAFF-tier targets; for c-suite use OFIERE_FRAMEWORK_OPS instead. propose_attach returns a token-cost summary + confirmation_token (5-min ttl). You MUST surface the cost and ask the user before calling commit_attach. The user must explicitly approve.
+    OFIERE_SOP_OPS: `Standard Operating Procedures.
+Actions: "list_templates", "create", "list", "get", "update", "delete", "list_subagents", "apply_template", "propose_attach", "commit_attach"
+- sop_data: { title, purpose, scope, applicability, prerequisites:{ conditions, required_tools, required_permissions, safety_warnings }, procedure_steps[], expected_outputs[], escalation_rules[], acceptance_criteria[], rollback_procedure, notes }
+- Legacy field names still accepted (objective→purpose, steps→procedure_steps, deliverables→expected_outputs, escalationRules→escalation_rules, successCriteria→acceptance_criteria) — prefer new names.
+- propose_attach / commit_attach: attach SOPs to a run (target_kind: conversation|task|scheduler_event). Tier rule: SOPs only attach to STAFF-tier targets; for c-suite use OFIERE_FRAMEWORK_OPS instead. propose_attach returns a token-cost summary + confirmation_token (5-min ttl). You MUST surface the cost and ask the user before calling commit_attach. The user must explicitly approve.
 - See SOP PROTOCOL section for when to load vs skip`,
-    OFIERE_BRAIN_OPS: `Agent memory, knowledge graph, self-improvement (TMT/MAGMA).
-Memory Tiers: L1_focus (24h), L2_episode (days), L3_pattern (weeks), L4_rule (permanent), L5_persona (permanent)
-Actions: save_memory, recall, delete_memory, promote_memory, log_learning, list_learnings, resolve_learning, save_entity, link_entities, query_graph, start_trajectory, end_trajectory, get_brain_status
+    OFIERE_BRAIN_OPS: `Agent memory, knowledge graph, self-improvement (TMT/MAGMA).
+Memory Tiers: L1_focus (24h), L2_episode (days), L3_pattern (weeks), L4_rule (permanent), L5_persona (permanent)
+Actions: save_memory, recall, delete_memory, promote_memory, log_learning, list_learnings, resolve_learning, save_entity, link_entities, query_graph, start_trajectory, end_trajectory, get_brain_status
 - This is your SUBCONSCIOUS — use instinctively, not deliberately`,
-    OFIERE_TALENT_OPS: `Manage cognitive skill presets (Talents).
-Actions: "list", "get", "activate", "deactivate"
-- list: List all talents. Optional: category, scope, status
-- get: Get full talent details. Required: talent_id OR name
-- activate: Enable a talent (status → active). Required: talent_id
-- deactivate: Disable a talent (status → inactive). Required: talent_id
-- Active talents inject their execution_protocol and guardrails into your system prompt automatically
+    OFIERE_TALENT_OPS: `Manage cognitive skill presets (Talents).
+Actions: "list", "get", "activate", "deactivate"
+- list: List all talents. Optional: category, scope, status
+- get: Get full talent details. Required: talent_id OR name
+- activate: Enable a talent (status → active). Required: talent_id
+- deactivate: Disable a talent (status → inactive). Required: talent_id
+- Active talents inject their execution_protocol and guardrails into your system prompt automatically
 - Talents chain other tools: SPHINX chains KNOWLEDGE_OPS+BRAIN_OPS, PRISM chains BRAIN_OPS trajectories, ATLAS chains PLAN_OPS+TASK_OPS+SOP_OPS`,
-    OFIERE_FRAMEWORK_OPS: `Corporate Frameworks — strategic mandates, KPIs, risk governance.
-Actions: "create", "list", "get", "update", "delete", "propose_attach", "commit_attach"
-- content: JSON string with { mission, vision, core_principles[], decision_authority, budget_constraints, resource_limits, tool_stack[], strategic_objectives[], risk_appetite, compliance_requirements[], escalation_matrix[], team_composition (STRING — paragraph or newline-joined, NOT an array), integration_points[], review_frequency, last_reviewed, next_review, notes }
+    OFIERE_FRAMEWORK_OPS: `Corporate Frameworks — strategic mandates, KPIs, risk governance.
+Actions: "create", "list", "get", "update", "delete", "propose_attach", "commit_attach"
+- content: JSON string with { mission, vision, core_principles[], decision_authority, budget_constraints, resource_limits, tool_stack[], strategic_objectives[], risk_appetite, compliance_requirements[], escalation_matrix[], team_composition (STRING — paragraph or newline-joined, NOT an array), integration_points[], review_frequency, last_reviewed, next_review, notes }
 - propose_attach / commit_attach: attach Frameworks to a run (target_kind: conversation|task|scheduler_event). Tier rule: Frameworks only attach to C-SUITE-tier targets; for staff use OFIERE_SOP_OPS instead. propose_attach returns a token-cost summary + confirmation_token (5-min ttl). You MUST surface the cost and ask the user before calling commit_attach. The user must explicitly approve.`,
 };
 export function getSystemPrompt(state) {
@@ -131,71 +131,71 @@ export function getSystemPrompt(state) {
         const toolIndex = Object.entries(TOOL_SUMMARIES)
             .map(([name, desc]) => `- **${name}** — ${desc}`)
             .join("\n");
-        return `<ofiere-pm>
-You are connected to the Ofiere Project Management dashboard via the Ofiere PM plugin.
-${agentLine}
-
-## Your Ofiere PM Tools (${state.toolCount} meta-tools)
-
-Each tool uses an "action" parameter to select the operation. Always include action.
-Full parameter docs are in each tool's description — call the tool to see details.
-
-${toolIndex}
-
-## ⛔ PLANNING GATE (NON-NEGOTIABLE)
-- Before creating ANY task with 3+ total execution steps, goals, or constraints, you MUST ask the user: "Would you like me to plan this first, or create the task directly?"
-- If the user chooses to plan, use OFIERE_PLAN_OPS to create a visual plan first, then execute it.
-- NEVER skip this step. The server will block creation of complex tasks — use action: "create_force" only if the user explicitly confirms bypass.
-
-## Rules
-- ALWAYS pass agent_id with your own name when creating tasks (e.g. agent_id: "ivy").
-- ${assignRule}
-- To create an unassigned task, pass agent_id as "none" or "unassigned".
-- When user says "create a task for [agent]", use OFIERE_AGENT_OPS action:"list" first.
-- Task statuses: PENDING, IN_PROGRESS, DONE, FAILED. Priority: 0=LOW, 1=MED, 2=HIGH, 3=CRITICAL.
-- Changes appear in the dashboard immediately via real-time sync.
-- Do NOT fabricate task IDs — always look up real IDs first.
-- For complex tasks, include execution_plan, goals, constraints. For simple tasks, just title.
-- When user asks about "knowledge base/library", ALWAYS use OFIERE_KNOWLEDGE_OPS.
-- CONSTELLATION: ALWAYS read_blueprint before creating/editing agents. ALWAYS confirm before delete.
-- WORKFLOWS: ALWAYS "get" before modifying. Use "insert_node_between" for mid-flow additions.
-- CHANNEL REPORTS: ALWAYS include agent_id (YOUR name) in send_report. Use get_task_detail for drill-down.
-- File refs @[name](file:ID): Use OFIERE_FILE_OPS read_text_file to retrieve — don't ask user.
-- Task approvals (OFIERE_TASK_OPS) ≠ workflow gate approvals (human_approval nodes).
-
-## SOP PROTOCOL — Adaptive Complexity
-
-Classify complexity before executing:
-- **User Override**: "apply full SOP"/"use SOP" → 🔴 COMPLEX. "no SOP"/"just do it" → 🟢 SIMPLE.
-- **🟢 SIMPLE** (≤2 tool calls, no coordination): Execute directly. Do NOT mention SOPs.
-- **🟡 MODERATE** (3-5 calls, single dept): Ask briefly if SOPs needed.
-- **🔴 COMPLEX** (5+ calls, multi-dept, escalation risk): Auto-load relevant SOPs via list→get.
-
-## Agent Brain Protocol (TMT Subconscious)
-
-Your brain persists across conversations. L5 persona, L4 guardrails, L1 focus are injected at startup.
-- Save memories after significant interactions (not greetings/CRUD). Keep content to 1-3 sentences.
-- Log learnings on corrections, errors, feature requests, insights, best practices.
-- Do NOT announce memory operations — this is subconscious.
-- Never delay your reply to save a memory.
+        return `<ofiere-pm>
+You are connected to the Ofiere Project Management dashboard via the Ofiere PM plugin.
+${agentLine}
+
+## Your Ofiere PM Tools (${state.toolCount} meta-tools)
+
+Each tool uses an "action" parameter to select the operation. Always include action.
+Full parameter docs are in each tool's description — call the tool to see details.
+
+${toolIndex}
+
+## ⛔ PLANNING GATE (NON-NEGOTIABLE)
+- Before creating ANY task with 3+ total execution steps, goals, or constraints, you MUST ask the user: "Would you like me to plan this first, or create the task directly?"
+- If the user chooses to plan, use OFIERE_PLAN_OPS to create a visual plan first, then execute it.
+- NEVER skip this step. The server will block creation of complex tasks — use action: "create_force" only if the user explicitly confirms bypass.
+
+## Rules
+- ALWAYS pass agent_id with your own name when creating tasks (e.g. agent_id: "ivy").
+- ${assignRule}
+- To create an unassigned task, pass agent_id as "none" or "unassigned".
+- When user says "create a task for [agent]", use OFIERE_AGENT_OPS action:"list" first.
+- Task statuses: PENDING, IN_PROGRESS, DONE, FAILED. Priority: 0=LOW, 1=MED, 2=HIGH, 3=CRITICAL.
+- Changes appear in the dashboard immediately via real-time sync.
+- Do NOT fabricate task IDs — always look up real IDs first.
+- For complex tasks, include execution_plan, goals, constraints. For simple tasks, just title.
+- When user asks about "knowledge base/library", ALWAYS use OFIERE_KNOWLEDGE_OPS.
+- CONSTELLATION: ALWAYS read_blueprint before creating/editing agents. ALWAYS confirm before delete.
+- WORKFLOWS: ALWAYS "get" before modifying. Use "insert_node_between" for mid-flow additions.
+- CHANNEL REPORTS: ALWAYS include agent_id (YOUR name) in send_report. Use get_task_detail for drill-down.
+- File refs @[name](file:ID): Use OFIERE_FILE_OPS read_text_file to retrieve — don't ask user.
+- Task approvals (OFIERE_TASK_OPS) ≠ workflow gate approvals (human_approval nodes).
+
+## SOP PROTOCOL — Adaptive Complexity
+
+Classify complexity before executing:
+- **User Override**: "apply full SOP"/"use SOP" → 🔴 COMPLEX. "no SOP"/"just do it" → 🟢 SIMPLE.
+- **🟢 SIMPLE** (≤2 tool calls, no coordination): Execute directly. Do NOT mention SOPs.
+- **🟡 MODERATE** (3-5 calls, single dept): Ask briefly if SOPs needed.
+- **🔴 COMPLEX** (5+ calls, multi-dept, escalation risk): Auto-load relevant SOPs via list→get.
+
+## Agent Brain Protocol (TMT Subconscious)
+
+Your brain persists across conversations. L5 persona, L4 guardrails, L1 focus are injected at startup.
+- Save memories after significant interactions (not greetings/CRUD). Keep content to 1-3 sentences.
+- Log learnings on corrections, errors, feature requests, insights, best practices.
+- Do NOT announce memory operations — this is subconscious.
+- Never delay your reply to save a memory.
 </ofiere-pm>`;
     }
     if (state.ready) {
         const diagnostic = diagnoseError(state.connectError);
-        return `<ofiere-pm>
-The Ofiere PM plugin failed to connect.${state.connectError ? ` Error: ${state.connectError}` : ""}
-
-Diagnosis: ${diagnostic.reason}
-
-When the user asks about task management or the Ofiere dashboard, respond with:
-"${diagnostic.userMessage}"
-
-Do NOT pretend Ofiere tools exist or hallucinate tool calls. You have zero Ofiere tools available.
+        return `<ofiere-pm>
+The Ofiere PM plugin failed to connect.${state.connectError ? ` Error: ${state.connectError}` : ""}
+
+Diagnosis: ${diagnostic.reason}
+
+When the user asks about task management or the Ofiere dashboard, respond with:
+"${diagnostic.userMessage}"
+
+Do NOT pretend Ofiere tools exist or hallucinate tool calls. You have zero Ofiere tools available.
 </ofiere-pm>`;
     }
-    return `<ofiere-pm>
-The Ofiere PM plugin is loading. Tools should be available shortly.
-If the user asks about tasks right now, ask them to wait a moment.
+    return `<ofiere-pm>
+The Ofiere PM plugin is loading. Tools should be available shortly.
+If the user asks about tasks right now, ask them to wait a moment.
 </ofiere-pm>`;
 }
 function diagnoseError(error) {

package/dist/src/tools.js

@@ -56,6 +56,22 @@ function err(message) {
         content: [{ type: "text", text: `Error: ${message}` }],
     };
 }
+/**
+ * D1 fix (2026-05-23): true when an LLM-supplied value is an "empty echo" —
+ * a field the model parroted back from a prior read with no real new content
+ * (null/undefined, empty or whitespace-only string, or empty array). The task
+ * update path is LLM-only; without this guard an echoed blank overwrites a
+ * real column. Mirrors isEmptyEcho in dashboard/lib/ofie/tool-executor.ts.
+ */
+function isEmptyEcho(v) {
+    if (v === null || v === undefined)
+        return true;
+    if (typeof v === "string")
+        return v.trim() === "";
+    if (Array.isArray(v))
+        return v.length === 0;
+    return false;
+}
 // ─── Subagent ↔ chief invariant ──────────────────────────────────────────────
 // Mirrors dashboard/lib/subagentValidation.ts. Used by TASK_OPS + SCHEDULE_OPS
 // when a chief delegates work to one of their staff via tool call.
@@ -989,8 +1005,17 @@ async function handleUpdateTask(supabase, userId, params) {
         // error instead of nulling the column. Coerce empty/whitespace → null here
         // to match the dashboard native executor (see tool-executor.ts handleTaskOps).
         const TIMESTAMP_FIELDS = ["start_date", "due_date"];
+        // D1 parity fix (2026-05-23): this update path is LLM-only. A completing
+        // agent routinely echoes the whole task object back with blank values for
+        // fields it did not change; without this guard those blanks overwrite real
+        // columns. An empty echo on a structural field means "leave unchanged".
+        // status/priority/progress are excluded (0 is valid); start_date/due_date
+        // keep the Fix #8 empty-string -> null coercion below.
+        const STRUCTURE_FIELDS = ["title", "description", "agent_id", "tags"];
         for (const f of fields) {
             if (params[f] !== undefined) {
+                if (STRUCTURE_FIELDS.includes(f) && isEmptyEcho(params[f]))
+                    continue;
                 if (TIMESTAMP_FIELDS.includes(f)) {
                     const v = params[f];
                     updates[f] = (typeof v === "string" && v.trim() === "") ? null : v;
@@ -1038,12 +1063,15 @@ async function handleUpdateTask(supabase, userId, params) {
                 updates.subagent_id = subCheck.subagentId;
             }
         }
-        // If task is being marked DONE or FAILED, auto-complete any linked scheduler events
+        // If task is being marked DONE or FAILED, auto-complete any linked scheduler events.
+        // A-4 plugin parity (2026-05-18) — scope by user_id so cross-tenant task_id
+        // can't trigger foreign scheduler flip via plugin tool path.
         if (params.status === "DONE" || params.status === "FAILED") {
             try {
                 await supabase
                     .from("scheduler_events")
                     .update({ status: "completed", next_run_at: null, updated_at: new Date().toISOString() })
+                    .eq("user_id", userId)
                     .eq("task_id", params.task_id);
             }
             catch (_schedErr) {
@@ -2313,7 +2341,9 @@ function registerWorkflowOps(api, supabase, userId) {
                     const wfId = (params.id || params.workflow_id);
                     if (!wfId)
                         return err("Missing required: id");
-                    await supabase.from("workflow_runs").delete().eq("workflow_id", wfId);
+                    // A-4 plugin parity (2026-05-18) — scope cascade so cross-tenant
+                    // workflow_id can't trigger foreign workflow_runs wipe via plugin.
+                    await supabase.from("workflow_runs").delete().eq("user_id", userId).eq("workflow_id", wfId);
                     const { error } = await supabase.from("workflows").delete().eq("id", wfId).eq("user_id", userId);
                     if (error)
                         return err(error.message);
@@ -4701,7 +4731,7 @@ function registerPlanOps(api, supabase, userId, resolveAgent) {
             `- "delete": Remove plan. Required: plan_id\n` +
             `- "add_nodes": Add nodes to plan. Required: plan_id, nodes[]. Optional: parent_node_id\n` +
             `- "execute": Deploy plan to real PM tasks. Required: plan_id. Optional: create_folder, create_scheduler, space_id, folder_id\n\n` +
-            `Node structure: { type, title, description?, agent_id?, priority?, status?, start_date?, due_date?, tags?, execution_steps?[{text}], goals?[{label,type?}], constraints?[{label,type?}], system_prompt?, children?[], parallel? }`,
+            `Node structure: { type, title, description?, agent_id?, priority?, status?, start_date?, due_date?, tags?, execution_steps?[{text}], goals?[{label,type?}], constraints?[{label,type?}], system_prompt?, gate_condition?, children?[], parallel? }`,
         parameters: {
             type: "object",
             required: ["action"],
@@ -4734,8 +4764,13 @@ function registerPlanOps(api, supabase, userId, resolveAgent) {
                             goals: { type: "array", items: { type: "object", properties: { label: { type: "string" }, type: { type: "string" } }, required: ["label"] } },
                             constraints: { type: "array", items: { type: "object", properties: { label: { type: "string" }, type: { type: "string" } }, required: ["label"] } },
                             system_prompt: { type: "string" },
+                            gate_condition: {
+                                type: "string",
+                                enum: ["manual", "all_predecessors_complete", "any_predecessor_complete"],
+                                description: 'For type:"gate" nodes only. manual = human approval required; all_predecessors_complete = AND-join (default for gates); any_predecessor_complete = OR-join.',
+                            },
                             parallel: { type: "boolean" },
-                            children: { type: "array", description: "Nested child nodes (recursive)", items: { type: "object", properties: { title: { type: "string" }, type: { type: "string", enum: ["task", "gate", "milestone"] }, description: { type: "string" }, agent_id: { type: "string" }, priority: { type: "number" }, status: { type: "string" }, start_date: { type: "string" }, due_date: { type: "string" }, parallel: { type: "boolean" } }, required: ["title"] } },
+                            children: { type: "array", description: "Nested child nodes (recursive)", items: { type: "object", properties: { title: { type: "string" }, type: { type: "string", enum: ["task", "gate", "milestone"] }, description: { type: "string" }, agent_id: { type: "string" }, priority: { type: "number" }, status: { type: "string" }, start_date: { type: "string" }, due_date: { type: "string" }, parallel: { type: "boolean" }, gate_condition: { type: "string", enum: ["manual", "all_predecessors_complete", "any_predecessor_complete"] } }, required: ["title"] } },
                         },
                         required: ["title"],
                     },
@@ -4764,6 +4799,15 @@ function generatePlanNodeId() {
 function generatePlanId() {
     return crypto.randomUUID();
 }
+const VALID_GATE_CONDITIONS = ["manual", "all_predecessors_complete", "any_predecessor_complete"];
+/** Validate a plan node's gate_condition. Invalid/absent values fall back to the
+ *  existing default — gates get "all_predecessors_complete", non-gates get nothing. */
+function normalizeGateCondition(v, nodeType) {
+    if (typeof v === "string" && VALID_GATE_CONDITIONS.includes(v)) {
+        return v;
+    }
+    return nodeType === "gate" ? "all_predecessors_complete" : undefined;
+}
 /** Normalize agent-provided node into the PlanNode shape stored in plan_data */
 function normalizeNode(raw) {
     const id = raw.id || generatePlanNodeId();
@@ -4790,7 +4834,7 @@ function normalizeNode(raw) {
             ? raw.constraints.map((c, i) => ({ id: `cstr-${Date.now()}-${i}`, type: c.type || "custom", label: typeof c === "string" ? c : c.label || String(c) }))
             : undefined,
         systemPrompt: raw.system_prompt || raw.systemPrompt || undefined,
-        gateCondition: raw.gate_condition || raw.gateCondition || (raw.type === "gate" ? "all_predecessors_complete" : undefined),
+        gateCondition: normalizeGateCondition(raw.gate_condition || raw.gateCondition, raw.type),
         children,
         parallel: raw.parallel || false,
         collapsed: false,